PERFORMANCE ANALYSIS OF VARIOUS IMAGE CLASSIFIERS AND FORMULATION OF CONVOLUTION LAYER OF CNN

Abstract

Convolutional Neural Networks (CNNs) are a kind of deep neural networks which are designed from the biologically driven models. Researchers focused on how human perceives an image in their brain. As image is passed through different layers in human brain, in the same way CNNs have many layers. In this work, structure of CNN is described, along with the guidelines on the design of convolution layer and decision making on when to use pre-trained CNN model with transfer learning and when to design own custom architecture CNN model. This will help future researchers in a quick start with CNN modeling. Experimentation is done on two popular image datasets i.e., CIFAR-100 and Stanford Clothing Attribute Dataset, where CIFAR -100 is a clean dataset of 60,000 images belonging to 100 classes and Stanford Clothing Attribute Dataset is highly noisy and imbalanced data as it has uneven distribution of samples for different attributes and many of the samples do not have a clear distinction between the classes resulting in overlapping training data.

To display the results, four CNNs were designed, where two models were pre-trained CNN models and two were customized CNN models. A comparative analysis of their performance on image classification task and treatment of the missing data in dataset is being done. Based on this comparison and related study, the guidelines for designing convolution layer and making choice between using a pre-trained (transfer learning) or customized CNNs are made. Along with this work, the performance of various machine learning classifiers such as Multinomial Logistic Regression, Support Vector Machine, Multi-Layer Perceptron, Random Forests, Naïve Bayes, K- Nearest Neighbors, ADA Boost and Convolutional Neural Network is compared over two popular image data sets CIFAR-10 and MNIST. The analysis is done based on performance variance, feature extraction and feature selection. The results show that CNNs outperformed amongst all the image classifiers by automatically extracting and selecting features and giving better results.